Hi Hunters! 👋 Rémi here, maker of OpenWhistle. I built OpenWhistle because every team I've worked with — sales, support, HR — was leaving the same gap on the floor: voice conversations never become structured data. Notes get scribbled, CRMs stay empty, action items live in Slack DMs, and a week later nobody remembers what was agreed. The fix shouldn't be "manually log every call". The fix is: define what you want to extract once, then drop the audio. The agent does the rest — transcribes (Mistral Voxtral or local Whisper), runs a hybrid extractor (deterministic rules first, LLM fills the gaps), and emits a JSON payload + a Markdown report keyed exactly on the fields you declared. Ready for your CRM, your ticket system, your Slack channel. A few things I'm proud of: Hybrid extraction. Most competitors are "throw the transcript at GPT and hope". We run regex + enum normalization + numeric coercion first, the LLM only fills the holes. Predictable cost, predictable latency, accuracy that scales with field difficulty. AI schema drafter. You describe the data you want in plain English, Mistral medium-3.1 returns a strictly-validated plan in our dialect. Beats hand-writing JSON Schema. Live in the agent builder under the "AI" tab. Three-stage STT resilience. Public marketplace + share links. Flip an agent to public, anyone can install with one click. Or mint a signed share link with max_uses + TTL. EU-hosted, EU-built, GDPR-native. No US-only LLM dependency. Free to start (30 seconds of audio, no credit card) I'll be camped on this thread all day — drop your toughest "but what about X" and I'll answer with code links. 🎙️

OpenWhistle

Audio in. Structured intelligence out.

Audio in. Structured intelligence out.